Advanced Co-Simulation and Neural Network based DPD Function Modeling RF Power Amplifiers

Power Amplifiers (PAs) are crucial components in modern communication systems, responsible for amplifying signals to ensure effective transmission. However, PAs introduce distortion, which can degrade signal quality and system performance. Accurate modeling and simulation of PAs are essential for designing efficient and reliable communication systems.

Today, PA design and modeling are often performed using a combination of circuit-level simulations and empirical data in different software environments and time phase of the design process. This can lead to inefficiencies and inaccuracies in the modeling process and slow down the design cycle.

In addition, PA architectures are becoming increasingly complex, and traditional modeling techniques may not be sufficient to capture the nonlinear behavior of PAs accurately. Recurrent Neural Networks (RNNs) have shown promise in modeling the behavior of PAs, but there is still room for improvement.

This internship project aims to explore the state-of-the-art in PA modeling and Digital Predistortion (DPD) design. The project focuses on two main tasks: - automating a co-simulation workflow that integrates Advanced Design System (ADS) and Python for PA/DPD modeling; - exploring innovative approaches in simulation-based DPD function design.

Currently, the PA design process involves several manual steps: running ADS simulations, exporting voltages from the ADS simulation bench, executing Python scripts, manually exploring DPD models, generating predistorted signals, and starting ADS simulations for validation with these signals. Automating this process will make it more efficient and reduce errors. One option is to push the project further by looking into identifying memory polynomial hyper-parameters from harmonic balance and two-tone analysis.

For the investigation of new DPD functions, the project offer different axes like the use of complex-valued neural networks and the piecewise continuity modeling to better capture the nonlinear behavior of PAs. Complex-valued neural networks can handle complex inputs and outputs, which is useful for modeling communication signals. However, training these networks is challenging due to the complexity of the operations and the need for suitable activation functions. Piecewise continuity modeling involves dividing the PA behavior into segments and modeling each segment separately, which can improve accuracy but requires careful management of transitions between segments.

Another challenge is handling unseen PAPR cases. PAPR (Peak-to-Average Power Ratio) is important in PA design because high PAPR signals can cause significant distortion. Modeling and simulating PAs for unseen PAPR cases requires robust and adaptive models that can generalize well to new data. This involves developing techniques to handle varying signal characteristics and ensuring that the models remain accurate and reliable under different operating conditions.

A geat effort will be put on making the project reproducible and shareable. This includes using version control systems like git, documenting the code and processes thoroughly, and sharing the results in a way that others can easily understand and replicate for example through an open-source repository.

This project will give students the opportunity to work on cutting-edge research in the field of power amplifier modeling and digital predistortion. Students will gain experience with advanced simulation tools and machine learning techniques.

This project requires a good knowledge of machine learning concepts, telecom systems, signal processing, and programming.

Please send your application to the internship supervisor (please see headings). Your application should include :

Deadline for application: 1st March 2024.

Upon reception of your application, we will contact you for an interview. The interview agenda is usually as follows :

Literrature review will focus on subitted (yet unpublished) papers from the team: